For autonomous underwater vehicle (AUV) actuator faults, a fault diagnosis method based on Gaussian particle filter was proposed. First,  parameters were introduced to express the control action loss of each degree caused by actuator failures and a fault model was built based on the equation of motion in sixdegrees of freedom to describe the actuator fault. Secondly,  these arguments were estimated in company with the motion states by using the modified Gaussian particle filter. Thirdly, those parameters were used to detect faults using modified Bayesian and estimate the amplitude of the failure with sliding window. Finally, simulation was conducted and some data from sea experiment were analyzed to test the algorithm. The results show that the proposed methods can detect and diagnose faults fast and accurately.